Our goal for the first two weeks

Transcription

Our goal for the first two weeks
Our goal for the first two weeks
. . . is basically to get comfortable using R. We will learn
• to do some elementary statistics
Statistics with R
Introduction and Examples
• to use the documentation / help system
• about the language basics
Deepayan Sarkar
• about data manipulation
University of Wisconsin – Madison
Summer Institute for Training in Biostatistics (2005)
What is R
• From the text: R provides an environment in which you
can perform statistical analysis and produce graphics. It is
actually a complete programming language, although that is
only marginally described in this book.
• We will mostly use R as a toolbox for standard statistical
techniques.
We will learn about other specialized tools later when they are
required.
Interacting with R
R usually works interactively, using a question-and-answer model:
• Start R
• Type a command and press Enter
• R executes this command (often printing the result)
• Some knowledge of R programming essential to use it well.
• R then waits for more input
More information available at the R Project homepage:
http://www.r-project.org
• Type q() to exit
Simple Examples
Variables
> 2 + 2
[1] 4
> exp(-2)
[1] 0.1353353
> log(100, base = 10)
[1] 2
> runif(10)
[1] 0.46164277 0.84000664 0.34755931 0.77642092 0.37276779
[6] 0.23124987 0.08789971 0.60120298 0.38273254 0.16632850
The last command generates 10 random numbers between 0 and 1. The
printed result is a vector of 10 numbers.
The bracketed numbers at the
beginning of each line indicate the index of the first number on that line.
exp, log and runif are functions, indicated by the presence of
parentheses. Most useful things in R are done by functions.
Like most programming languages, R has symbolic variables
which can be assigned values. The traditional way to do this
in R is the ’<-’ operator. (The more common ’=’ also works.)
> x <- 2
> yVar2 = x + 3
> s <- "this is a character string"
> x
[1] 2
> yVar2
[1] 5
> s
[1] "this is a character string"
> x + x
[1] 4
> x^yVar2
[1] 32
Variables
> plot(runif(10))
Possible variables names are very flexible. However, note that
1.0
Simple Examples: plotting
• variable names cannot start with a digit
0.9
●
• names are case-sensitive
●
0.6
0.7
●
0.5
●
●
0.4
●
●
0.3
runif(10)
0.8
●
●
●
2
4
6
Index
8
10
• some common names are already used by R, e.g.,
c, q, t, C, D, F, I, T,
and should be avoided
Vectorized arithmetic
• The elementary data types in R are all vectors
• The c(...) construct can be used to create vectors:
> weight <- c(60, 72, 57, 90, 95, 72)
> weight
[1] 60 72 57 90 95 72
• To generate a vector of regularly spaced numbers, use
Scalars from Vectors
Many functions summarize a data vector by producing a scalar
from a vector. For example
> sum(weight)
[1] 446
> length(weight)
[1] 6
> avg.weight <- mean(weight)
> avg.weight
[1] 74.33333
> seq(0, 1, length = 11)
[1] 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
> 1:10
[1] 1 2 3 4 5 6 7 8 9 10
• Common arithmetic operations (including +, -, *, /, ^) and
mathematical functions (e.g. sin, cos, log) work elementwise on vectors, and produce another vector:
> 1:10 + c(0, 2)
[1] 1 4 3 6
5
8
7 10
9 12
> plot(x = height, y = weight)
90
●
80
●
●
70
When two vectors are not of equal length, the shorter one is
recycled. The following adds 0 to all the odd elements and 2 to
all the even elements of 1:10:
The simplest way to produce R graphics output is to use the
plot function:
60
> height <- c(1.75, 1.8, 1.65, 1.9, 1.74, 1.91)
> height^2
[1] 3.0625 3.2400 2.7225 3.6100 3.0276 3.6481
> bmi <- weight/height^2
> bmi
[1] 19.59184 22.22222 20.93664 24.93075 31.37799 19.73630
> log(bmi)
[1] 2.975113 3.101093 3.041501 3.216102 3.446107 2.982460
Graphics
weight
Vectorized arithmetic
●
●
●
1.65
1.70
1.75
1.80
height
1.85
1.90
Graphics
Descriptive Statistics (contd)
There are many options that can control the details of what the
plot looks like. Once created, plots can also be enhanced:
Simple summary statistics: quantiles, inter-quartile range
>
>
>
>
plot(x = height, y = weight, pch = 2, col = "red")
hh <- c(1.65, 1.7, 1.75, 1.8, 1.85, 1.9)
lines(x = hh, y = 22.5 * hh^2)
title(main = "weight vs height, with BMI = 22.5 curve")
80
70
60
weight
90
weight vs height, with BMI = 22.5 curve
1.65
1.70
1.75
1.80
1.85
1.90
> xquants <- quantile(x)
> xquants
0%
25%
50%
75%
100%
-2.3011444 -0.9792947 -0.3380632 0.4119879 2.8260118
> xquants[4] - xquants[2]
75%
1.391283
> IQR(x)
[1] 1.391283
> quantile(x, probs = seq(0, 1, length = 11))
0%
10%
20%
30%
40%
-2.30114444 -1.55916781 -1.12134696 -0.77464565 -0.61983638
50%
60%
70%
80%
90%
-0.33806323 -0.01043648 0.16151194 0.45262367 0.82814147
100%
2.82601184
height
Descriptive Statistics
The summary function
Simple summary statistics: mean, median, standard deviation,
variance
When applied to a numeric vector, summary produces a nice
summary display:
> x <- rnorm(100)
> mean(x)
[1] -0.2881313
> sd(x)
[1] 0.983495
> var(x)
[1] 0.9672625
> median(x)
[1] -0.3380632
> summary(x)
Min. 1st Qu. Median
Mean 3rd Qu.
-2.3010 -0.9793 -0.3381 -0.2881 0.4120
Max.
2.8260
The output of summary can be different when applied to other
objects.
The Iris Data Set
Graphical display: Strip Plots
Let’s look at a real example: The Iris data set is one of many
already available in R (type data() for a full list).
The simplest plot of numeric data is a strip plot
(often called a dot plot)
> str(iris)
‘data.frame’:
$ Sepal.Length:
$ Sepal.Width :
$ Petal.Length:
$ Petal.Width :
$ Species
:
150 obs. of 5 variables:
num 5.1 4.9 4.7 4.6 5 5.4 4.6 5 4.4 4.9 ...
num 3.5 3 3.2 3.1 3.6 3.9 3.4 3.4 2.9 3.1 ...
num 1.4 1.4 1.3 1.5 1.4 1.7 1.4 1.5 1.4 1.5 ...
num 0.2 0.2 0.2 0.2 0.2 0.4 0.3 0.2 0.2 0.1 ...
Factor w/ 3 levels "setosa","versicolor",..: 1 1 1 1 1 1 1 1 1 1 ...
> stripchart(x)
This data set contains measurements on 150 flowers, 50 each
from 3 species: Iris setosa, versicolor and virginica.
It is typically used to illustrate the problem of classification—
given the four measurements for a new flower, can we predict
its Species?
The summary function revisited
> summary(iris)
Sepal.Length
Min.
:4.300
1st Qu.:5.100
Median :5.800
Mean
:5.843
3rd Qu.:6.400
Max.
:7.900
Petal.Width
Min.
:0.100
1st Qu.:0.300
Median :1.300
Mean
:1.199
3rd Qu.:1.800
Max.
:2.500
Sepal.Width
Min.
:2.000
1st Qu.:2.800
Median :3.000
Mean
:3.057
3rd Qu.:3.300
Max.
:4.400
Species
setosa
:50
versicolor:50
virginica :50
Petal.Length
Min.
:1.000
1st Qu.:1.600
Median :4.350
Mean
:3.758
3rd Qu.:5.100
Max.
:6.900
Note the different format of the output. Species is summarized
differently because it is a categorical variable (more commonly
called factor in R).
−2
−1
0
1
2
3
Strip Plots for the Iris data
We can produce a similar plot with the Iris data. Unfortunately,
it’s not possible to indicate which points came from which
Species.
> stripchart(iris[, 1:4], method = "stack", pch = 16,
+
cex = 0.4, offset = 0.6)
Petal.Width
Strip Plots for the Iris data
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Grouped Display
> library(package = "lattice")
> stripplot(type ~ measure, data = iris2, groups = Species,
+
jitter = TRUE, auto.key = list(columns = 3))
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
setosa
●
●
●
●
Petal.Length
Sepal.Width
Sepal.Width
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Sepal.Length
versicolor
virginica
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Sepal.Length
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
2
4
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
6
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●●
●●●●●●●
●●
●
●
●
●
●
● ●
● ●●●●
●
●●
●●●
●
●
●
●●
●●
●
●●
●
●
●●
●●
● ●
●
●
●
●●
●
●
●
●
●●●●
Petal.Length
●
●
●
●
●
●
●
●
Petal.Width
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
● ●●●
● ●● ●
●●● ●
●●
●
●●●●
●
●●●● ●●
●
●●●
● ●●●
●
●●
●
●
●
●
●●●
●●
●
●●
●●
●
●
●
●●
●
●●
●
●●
●●
●●
●●●
●●
●●●
0
0
●
●
●
●
●
●
2
4
6
8
measure
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
8
Grouped Display
Histograms
A more suitable plotting function is available in an add-on
package called lattice, but that needs the data to be in a slightly
different structure:
Strip plots are pretty much useless for large data sets. The
most popular (mostly because they are easy to draw by hand)
graphical summary for numeric data is the histogram
> hist(x)
15
10
5
Frequency
20
Histogram of x
0
> iris2 <- reshape(iris, varying = list(names(iris)[1:4]),
+
v.names = "measure", timevar = "type", times = names(iris)[1:4],
+
direction = "long")
> str(iris2, give.attr = FALSE)
‘data.frame’:
600 obs. of 4 variables:
$ Species: Factor w/ 3 levels "setosa","versicolor",..: 1 1 1 1 1 1 1 1 1 1 ...
$ type
: chr "Sepal.Length" "Sepal.Length" "Sepal.Length" "Sepal.Length" ...
$ measure: num 5.1 4.9 4.7 4.6 5 5.4 4.6 5 4.4 4.9 ...
$ id
: int 1 2 3 4 5 6 7 8 9 10 ...
−2
−1
0
1
x
2
3
Density Plots
Grouped Display (Histogram)
Density plots are generalized histograms
> histogram(~measure | type * Species, iris2, nint = 25)
0
> plot(density(x))
Percent of Total
0.3
0.2
0.1
Density
4
6
8
0
2
4
6
virginica
Petal.Width
virginica
Sepal.Length
virginica
Sepal.Width
versicolor
Petal.Length
versicolor
Petal.Width
versicolor
Sepal.Length
versicolor
Sepal.Width
0.0
0
2
80
60
40
20
0
setosa
Petal.Length
setosa
Petal.Width
setosa
Sepal.Length
setosa
Sepal.Width
80
60
40
20
0
4
0
N = 100 Bandwidth = 0.3524
2
4
6
8
0
2
4
6
8
measure
Grouped Display (Density Plot)
Grouped Scatter Plot
Again, for grouped data, the analogous lattice functions are
more suitable.
> xyplot(Petal.Length ~ Petal.Width, iris, groups = Species,
+
aspect = 1, auto.key = list(side = "right"))
7
> densityplot(~measure | type, data = iris2, groups = Species,
+
scales = "free", plot.points = FALSE)
Petal.Width
Sepal.Length
6
Sepal.Width
Petal.Length
1.0
0.5
0.5
4
6
1.0
5
4
6
0.0 0.5 1.0 1.5 2.0 2.5
0.0
0.0
2
2
setosa
versicolor
virginica
4
3
0
Density
0.0 0.5 1.0 1.5 2.0 2.5
Petal.Length
4
measure
5
8
80
60
40
20
0
density.default(x = x)
−2
2
virginica
Petal.Length
6
7
8
2
2
3
●
● ●
●
● ● ●
● ● ●
● ●
●
●
●
4
1
0.0
●
● ●
●
●
●
●
0.5
1.0
1.5
Petal.Width
2.0
2.5
●
Categorical Data
Tables
We have already seen one example:
Association between categorical variables can be summarized by
tables:
> summary(iris$Species)
setosa versicolor virginica
50
50
50
Let’s try to predict the Species using other measurements.
• What’s the best measure to use?
• What are good thresholds?
Discretizing
A continuous measure can be converted into a categorical one
using the cut function:
> PL.tab <- xtabs(~PL.disc + Species, iris)
> SL.tab <- with(iris, table(SL.disc, Species))
> PL.tab
Species
PL.disc
setosa versicolor virginica
(0,2.5]
50
0
0
(2.5,4.75]
0
44
1
(4.75,7]
0
6
49
> SL.tab
Species
SL.disc
setosa versicolor virginica
(0,5.5]
47
11
1
(5.5,6.25]
3
25
12
(6.25,8]
0
14
37
Graphical Display of Tables: Bar chart
> par(mfrow = c(1, 2))
> barplot(PL.tab, beside = TRUE, main = "Petal Length")
> barplot(SL.tab, beside = TRUE, main = "Sepal Length")
Sepal Length
0
0
10
10
20
20
30
30
40
40
50
Petal Length
> iris$PL.disc <- cut(iris$Petal.Length, breaks = c(0,
+
2.5, 4.75, 7))
> iris$SL.disc <- cut(iris$Sepal.Length, breaks = c(0,
+
5.5, 6.25, 8))
> str(iris)
‘data.frame’:
150 obs. of 7 variables:
$ Sepal.Length: num 5.1 4.9 4.7 4.6 5 5.4 4.6 5 4.4 4.9 ...
$ Sepal.Width : num 3.5 3 3.2 3.1 3.6 3.9 3.4 3.4 2.9 3.1 ...
$ Petal.Length: num 1.4 1.4 1.3 1.5 1.4 1.7 1.4 1.5 1.4 1.5 ...
$ Petal.Width : num 0.2 0.2 0.2 0.2 0.2 0.4 0.3 0.2 0.2 0.1 ...
$ Species
: Factor w/ 3 levels "setosa","versicolor",..: 1 1 1 1 1 1 1 1 1 1 ...
$ PL.disc
: Factor w/ 3 levels "(0,2.5]","(2.5,4.75]",..: 1 1 1 1 1 1 1 1 1 1 ...
$ SL.disc
: Factor w/ 3 levels "(0,5.5]","(5.5,6.25]",..: 1 1 1 1 1 1 1 1 1 1 ...
setosa
versicolor
virginica
setosa
versicolor
virginica
Graphical Display of Tables: Dot chart
Titanic Survivors
> dotchart(SL.tab, xlab = "Sepal Length")
> mosaicplot(Titanic, color = TRUE)
Titanic
No
Child
1st
Adult
2nd
Child Adult
3rd
Adult
Child
Crew
Adult
Child
●
Male
Yes
●
●
versicolor
(6.25,8]
(5.5,6.25]
(0,5.5]
Sex
●
●
virginica
(6.25,8]
(5.5,6.25]
(0,5.5]
No
●
Female
Yes
setosa
(6.25,8]
(5.5,6.25]
(0,5.5]
●
●
●
0
10
20
30
40
Class
Sepal Length
Higher Dimensional Tables
Titanic Survivors (simplified)
The built-in Titanic data set is a cross-tabulation of 4 characteristics of 2201 passengers on the Titanic
> mosaicplot(apply(Titanic, c(1, 2, 4), sum), color = TRUE)
2nd
Yes
No
No
3rd
"Crew"
$Sex
[1] "Male"
"Female"
Sex
Male
> dimnames(Titanic)
$Class
[1] "1st" "2nd" "3rd"
apply(Titanic, c(1, 2, 4), sum)
1st
Yes
No
$Survived
[1] "No" "Yes"
Female
$Age
[1] "Child" "Adult"
Class
Yes
No
Crew
Yes
Getting help
Getting help (contd)
R has too many tools for anyone to remember them all, so it is
very important to know how to find relevant information using
the help system.
The help pages can be opened in a browser as well:
• help.start()
Starts a browser window with an HTML help interface. One of the best
ways to get started. Has links to a very detailed manual for beginners
called ‘An Introduction to R’, as well as topic-wise listings.
• help(topic)
Displays the help page for a particular topic or function. Every R function
has a help page.
> help(plot, htmlhelp = TRUE)
The help pages are usually very detailed. Among other things,
they often contain
• A ‘See Also’ section that lists related help pages
• A Description of what the function returns
• An ‘Examples’ section, with actual code illutrating how to use the
documented functions. These examples can actually be run directly using
the example function. e.g., try
> example(plot)
• help.search("search string")
Subject/keyword search
Getting help (contd)
R packages
The help function provides topic-wise help. When you know
which function you are interested in, this is usually the best way
to learn how to use it. There’s also a short-cut for this; use
a question mark (?) followed by the topic. The following are
equivalent:
R makes use of a system of packages (section 1.5.3).
> help(plot)
> ?plot
When you want to know about a specific subject, but don’t know
which particular help page has the information, the help.search
function is very useful. For example, try
> help.search("logarithm")
• Each package is a collection of routines with a common
theme
• The core of R itself is a package called base
• A collection of packages is called a library
• Some packages are already loaded when R starts up. Other
packages need be loaded using the library() function
R packages
R packages
Several packages come pre-installed with R.
Other packages can be loaded by the user. We will be interested
in the ISwR package, which contains the datasets used in the
text. This can be loaded by:
> rownames(installed.packages())
[1] "KernSmooth" "MASS"
"base"
[5] "class"
"cluster"
"datasets"
[9] "grDevices" "graphics"
"grid"
[13] "methods"
"mgcv"
"nlme"
[17] "rpart"
"spatial"
"splines"
[21] "stats4"
"survival"
"tcltk"
[25] "utils"
"boot"
"foreign"
"lattice"
"nnet"
"stats"
"tools"
There are also many (more than 300) other packages contributed
by various users of R available online, from the Comprehensive
R Archive Network (CRAN):
http://cran.us.r-project.org/src/contrib/PACKAGES.html
The Bioconductor project provides an extensive collection of R
packages specifically for bioinformatics
http://www.bioconductor.org/packages/bioc/stable/src/contrib/html/
R packages
It is fairly easy for anyone to write new R packages. This is one
of the attractions of R over other statistical software.
Some packages are already loaded when R starts up. At any
point, The list of currently loaded packages can be listed by the
search function:
> search()
[1] ".GlobalEnv"
[3] "package:tools"
[5] "package:stats"
[7] "package:grDevices"
[9] "package:datasets"
[11] "package:base"
"package:lattice"
"package:methods"
"package:graphics"
"package:utils"
"Autoloads"
> library(ISwR)
New packages can be downloaded and installed using the
install.packages function. For example, to install the ISwR
package (if it’s not already installed), one can use
> install.packages("ISwR")
> library(help = ISwR)
The last call gives a list of all help pages in the package.